Is the Big Techs' billion-dollar investment in AI actually worth it?

Por Arthur Bittencourt ·
The current market is observing an unprecedented movement, with the largest tech companies investing approximately $800 billion just in Artificial Intelligence (AI) infrastructure this year. This spending is not motivated by mere technological curiosity; it is a defensive strategy to protect established business models, such as Google ads or cloud services from Amazon and Microsoft. The core fear lies in the possibility that new automated tools could cannibalize already consolidated revenue.
In practice, this volume of capital guarantees the end-user quick access to increasingly sophisticated and affordable tools, but it creates extremely high barriers to entry for new competitors. While the consumer sees efficient virtual assistants integrated into their devices, the real cost is concentrated in the invisible "factories" of these corporations. These factories process every individual request through inference—a technical term used when a ready model executes a specific task, functioning as if the factory were baking a new, personalized cake every time someone asks a question.
The foundation of this operation is Nvidia chips, which cost around six thousand dollars to manufacture but are sold for between thirty and forty thousand dollars. A single rack of this hardware can cost between three and five million dollars, consuming electricity equivalent to 100 American homes simultaneously. To sustain these machines, companies build gigantic data centers that function like industrial furnaces; in the case of Meta, its new plant in Louisiana was designed to reach 5 gigawatts, allowing it to power thousands of these devices with constant energy.
A disruption recently emerged with DeepSeek, a Chinese research lab that demonstrated training methods nearly as effective as American ones for under six million dollars. This feat caused panic in the markets, leading semiconductor stocks to drop drastically in a single day. The scenario illustrates Jevons' Paradox: as a technology becomes more efficient or cheaper (like a better 'recipe'), the demand increases so much that it ends up stimulating an even greater total consumption of the basic resource—in this case, the raw computational power available in the factories.
A pergunta sobre adotar já ou esperar depende da escala do seu objetivo tecnológico. Para desenvolvedores e usuários comuns, a tendência é aproveitar ferramentas cada vez menores e locais, como modelos rodando diretamente em celulares através do Gemini do Google ou tecnologias similares da Apple. Já para quem busca infraestrutura própria, os riscos são maiores devido à rápida depreciação das peças e às limitações físicas de conexão elétrica. No momento, enquanto a capacidade produtiva for escassa, quem dominar as fábricas terá vantagem estratégica antes que a eficiência extrema torne esses ativos obsoletos demais para gerar lucro.